fix(workspace-server): #1644 — include auth_token in POST /workspaces 201 response #1669

Merged
devops-engineer merged 11 commits from fix-1644-workspace-create-returns-auth-token into main 2026-06-02 00:17:50 +00:00
7 changed files with 253 additions and 2 deletions
@@ -0,0 +1,124 @@
# Engineer-Agent Gitea Token Scope Runbook
## Symptom
Engineer-class agents (e.g. `agent-dev-a`, `agent-dev-b`) fail swarm-pull issue discovery or receive HTTP 403 when calling Gitea issue-list APIs, while PR review and repository API operations continue to work.
Typical failing call:
```bash
GET /api/v1/repos/molecule-ai/molecule-core/issues?state=open&labels=approved&limit=50
# => 403 Forbidden
```
Typical working calls (same token):
```bash
GET /api/v1/repos/molecule-ai/molecule-core/pulls?state=open&limit=50
POST /api/v1/repos/molecule-ai/molecule-core/pulls/1666/comments
# => 200 OK
```
## Root Cause
Gitea v1.22.6 routes issue-list under the `Issue` scope category (`routers/api/v1/api.go:1379-1491`), while PR routes live under repository/pull routing (`api.go:1278-1305`). The scope gate derives required read/write level from HTTP method (`api.go:309-313`), so `GET /issues?...` requires `read:issue`.
Engineer-class agent PATs were provisioned with repository and PR scopes but without `read:issue`, causing the asymmetric 403.
## Detection
1. **Agent-side**: swarm-pull workflow logs show `403 Forbidden` on issue enumeration but not on PR list/review.
2. **Platform-side**: Gitea access logs show `GET /repos/{owner}/{repo}/issues` returning 403 for the affected token.
3. **Reproduction** (from any workspace with a suspected token):
```bash
TOKEN=$(cat /configs/secrets.d/GITEA_TOKEN)
PLATFORM="https://git.moleculesai.app"
# Should succeed — confirms token is live
curl -s -o /dev/null -w "%{http_code}" \
-H "Authorization: token $TOKEN" \
"$PLATFORM/api/v1/user"
# Will 403 if the token lacks read:issue
curl -s -o /dev/null -w "%{http_code}" \
-H "Authorization: token $TOKEN" \
"$PLATFORM/api/v1/repos/molecule-ai/molecule-core/issues?state=open&limit=1"
```
## Immediate Fix
### Step 1: Issue fresh PATs with correct scopes
From a Gitea site-admin account (or via the Gitea web UI → Settings → Applications):
1. Navigate to the affected user's profile (e.g. `agent-dev-a`).
2. Go to **Settings → Applications → Generate New Token**.
3. Select scopes:
- `read:repository` (existing)
- `write:repository` (existing, if push is required)
- `read:issue` (**add this**)
- `write:issue` (add only if agents must comment/edit issues)
- `read:pull-request` / `write:pull-request` (existing)
- `read:comment` / `write:comment` (existing, if PR review is required)
4. Copy the plaintext token immediately — it is shown only once.
### Step 2: Update workspace secrets
For each affected engineer workspace, update the Gitea token secret:
```bash
# Via the platform API (admin auth required)
PLATFORM="https://agents-team.moleculesai.app"
ADMIN_TOKEN="<your-admin-token>"
WORKSPACE_ID="<affected-workspace-id>"
NEW_GITEA_TOKEN="<fresh-token-from-step-1>"
curl -X POST "$PLATFORM/workspaces/$WORKSPACE_ID/secrets" \
-H "Authorization: Bearer $ADMIN_TOKEN" \
-H "Content-Type: application/json" \
-d "{
\"GITEA_TOKEN\": \"$NEW_GITEA_TOKEN\"
}"
```
Restart the workspace so the runtime re-reads secrets:
```bash
curl -X POST "$PLATFORM/workspaces/$WORKSPACE_ID/restart" \
-H "Authorization: Bearer $ADMIN_TOKEN"
```
### Step 3: Smoke-test
From the restarted workspace, verify all three paths:
```bash
# 1. Issue list (the previously failing path)
curl -s -H "Authorization: token $GITEA_TOKEN" \
"https://git.moleculesai.app/api/v1/repos/molecule-ai/molecule-core/issues?state=open&labels=approved&limit=1" | jq '.[0].number'
# 2. PR list (should still work)
curl -s -H "Authorization: token $GITEA_TOKEN" \
"https://git.moleculesai.app/api/v1/repos/molecule-ai/molecule-core/pulls?state=open&limit=1" | jq '.[0].number'
# 3. Swarm-pull discovery (end-to-end)
# Trigger the agent's autonomous tick or delegate a task that enumerates open issues.
```
## Long-Term Fix
Update the **workspace secret injection path** that writes `/configs/secrets.d/GITEA_TOKEN` for engineer-class agents. The provisioning template or secret-distribution job should request `read:issue` (and optionally `write:issue`) at token-creation time.
File locations to audit:
- `.gitea/scripts/` — any token-provisioning automation
- `infra/terraform/` or equivalent — IAM/secret-manager templates
- `workspace-configs-templates/` — engineer-class workspace templates that declare required secrets
## Prevention
1. **Token scope checklist**: when provisioning new engineer-class agent tokens, verify the scope set includes `read:issue` before distributing the secret.
2. **Monitoring**: add an agent health-check that probes `GET /repos/molecule-ai/molecule-core/issues?limit=1` and surfaces a non-fatal warning if it returns 403.
3. **Documentation**: update the onboarding runbook for new engineer agents to include the full required scope list.
## References
- Gitea issue #1750: [RCA: engineer-token read:issue scope gap blocks swarm-pull workflow](https://git.moleculesai.app/molecule-ai/molecule-core/issues/1750)
- Gitea source: `routers/api/v1/api.go:309-313` (scope gate), `api.go:1278-1305` (PR routing), `api.go:1379-1491` (issue routing)
- Related: PR #1542 (provisioner git-creds injection), PR #1669 (auth_token inline mint)
@@ -43,6 +43,8 @@ func TestWorkspaceCreate_WithParentID(t *testing.T) {
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO structure_events").
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO workspace_auth_tokens").
WillReturnResult(sqlmock.NewResult(0, 1))
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
@@ -79,6 +81,8 @@ func TestWorkspaceCreate_ExplicitClaudeCodeRuntime(t *testing.T) {
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO structure_events").
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO workspace_auth_tokens").
WillReturnResult(sqlmock.NewResult(0, 1))
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
@@ -300,6 +304,8 @@ func TestWorkspaceCreate_MaxConcurrentTasksOverride(t *testing.T) {
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO structure_events").
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO workspace_auth_tokens").
WillReturnResult(sqlmock.NewResult(0, 1))
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
@@ -384,6 +384,8 @@ func TestWorkspaceCreate(t *testing.T) {
// Expect RecordAndBroadcast INSERT for WORKSPACE_PROVISIONING
mock.ExpectExec("INSERT INTO structure_events").
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO workspace_auth_tokens").
WillReturnResult(sqlmock.NewResult(0, 1))
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
@@ -420,6 +422,76 @@ func TestWorkspaceCreate(t *testing.T) {
}
}
// TestWorkspaceCreate_ReturnsAuthToken_201 pins the inline-auth_token
// behaviour added for #1644. Pre-fix, the 201 response was
// {id, status, awareness_namespace, workspace_access} — callers had to
// make a separate POST to /admin/workspaces/:id/tokens (AdminAuth-gated,
// path-prefix differs in CP-admin deploys) OR fall back to the dev-only
// GET /admin/workspaces/:id/test-token (deliberately 404s on
// MOLECULE_ENV=production per feedback_no_dev_only_routes_in_e2e).
//
// Post-fix: every Create response includes an `auth_token` field with
// the freshly-minted plaintext bearer (returned once, never recoverable).
// This is the SSOT path — production E2E + canvas + org_import all
// get the bearer they need in the same round trip.
//
// Failure path is non-fatal: if the IssueToken DB call fails, the 201
// still goes out without auth_token + a fallback log line. That branch
// is exercised by sqlmock returning a non-INSERT-INTO-workspace_auth_tokens
// path here — the test asserts presence on the happy path.
func TestWorkspaceCreate_ReturnsAuthToken_201(t *testing.T) {
mock := setupTestDB(t)
setupTestRedis(t)
broadcaster := newTestBroadcaster()
handler := NewWorkspaceHandler(broadcaster, nil, "http://localhost:8080", "/tmp/configs")
mock.ExpectBegin()
mock.ExpectExec("INSERT INTO workspaces").
WithArgs(sqlmock.AnyArg(), "Token Holder", nil, 3, "claude-code", (*string)(nil), nil, "none", (*int64)(nil), models.DefaultMaxConcurrentTasks, "push").
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectCommit()
mock.ExpectExec("INSERT INTO canvas_layouts").
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO structure_events").
WillReturnResult(sqlmock.NewResult(0, 1))
// The inline mint added in #1644 Part B — wsauth.IssueToken issues
// a new bearer via INSERT INTO workspace_auth_tokens (workspace_id,
// token_hash, prefix). This is the assertion that the new code path
// reaches the DB.
mock.ExpectExec("INSERT INTO workspace_auth_tokens").
WithArgs(sqlmock.AnyArg(), sqlmock.AnyArg(), sqlmock.AnyArg()).
WillReturnResult(sqlmock.NewResult(0, 1))
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
body := `{"name":"Token Holder","model":"anthropic:claude-opus-4-7"}`
c.Request = httptest.NewRequest("POST", "/workspaces", bytes.NewBufferString(body))
c.Request.Header.Set("Content-Type", "application/json")
handler.Create(c)
if w.Code != http.StatusCreated {
t.Fatalf("expected 201, got %d: %s", w.Code, w.Body.String())
}
var resp map[string]interface{}
if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
t.Fatalf("parse response: %v", err)
}
tok, ok := resp["auth_token"].(string)
if !ok || tok == "" {
t.Fatalf("expected non-empty auth_token in 201 response (the #1644 SSOT inline mint), got: %s", w.Body.String())
}
// Sanity: tokens are base64-RawURL encoded 32-byte payloads (per
// wsauth/tokens.go::tokenPayloadBytes), so a meaningful lower bound
// is ~40 chars. If this fails, IssueToken's contract drifted.
if len(tok) < 40 {
t.Errorf("auth_token suspiciously short (%d chars) — wsauth.IssueToken contract drift?", len(tok))
}
if err := mock.ExpectationsWereMet(); err != nil {
t.Errorf("unmet sqlmock expectations — inline mint path may have skipped IssueToken: %v", err)
}
}
func TestBuildProvisionerConfig_WorkspacePathFromPayload(t *testing.T) {
setupTestDB(t)
broadcaster := newTestBroadcaster()
@@ -807,11 +807,38 @@ func (h *WorkspaceHandler) Create(c *gin.Context) {
}
}
c.JSON(http.StatusCreated, gin.H{
// Mint the workspace's first bearer token and return it inline
// (#1644). Pre-fix, callers had to make a separate POST to
// /admin/workspaces/:id/tokens (production path, AdminAuth-gated,
// but the path-prefix differs in CP-admin deploys so staging E2E
// got HTML 404) OR fall back to GET /admin/workspaces/:id/test-token
// (dev-only — deliberately 404s on MOLECULE_ENV=production per
// admin_test_token.go::TestTokensEnabled, which violates
// feedback_no_dev_only_routes_in_e2e). Inlining the first token here
// makes the create response the SSOT — every caller (canvas Save,
// org_import, E2E, third-party API) gets the bearer they need to
// authenticate /activity, /a2a, /memory etc. without an extra
// round trip to a separate mint endpoint.
//
// Failure is non-fatal: the workspace row already committed; the
// operator can recover via POST /admin/workspaces/:id/tokens
// (canonical admin mint) or POST /workspaces/:id/external/rotate
// (already-used for the external pre-register path above). We log
// the failure and return 201 without the field — callers that need
// the token will get a clear-shaped fallback (auth_token absent
// from response = use the admin mint path).
resp := gin.H{
"id": id,
"status": "provisioning",
"workspace_access": workspaceAccess,
})
}
if authToken, tokErr := wsauth.IssueToken(ctx, db.DB, id); tokErr != nil {
log.Printf("Create workspace %s: inline auth_token mint failed (non-fatal — caller can use POST /admin/workspaces/:id/tokens): %v", id, tokErr)
} else {
resp["auth_token"] = authToken
}
c.JSON(http.StatusCreated, resp)
}
// addProvisionTimeoutMs decorates a workspace response map with the
@@ -168,6 +168,8 @@ func TestWorkspaceBudget_Create_WithLimit(t *testing.T) {
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO structure_events").
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO workspace_auth_tokens").
WillReturnResult(sqlmock.NewResult(0, 1))
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
@@ -91,6 +91,8 @@ func TestWorkspaceCreate_WithCompute_PersistsComputeJSON(t *testing.T) {
mock.ExpectCommit()
mock.ExpectExec("INSERT INTO canvas_layouts").
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO workspace_auth_tokens").
WillReturnResult(sqlmock.NewResult(0, 1))
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
@@ -390,6 +390,8 @@ func TestWorkspaceCreate_DefaultsApplied(t *testing.T) {
// Expect RecordAndBroadcast INSERT
mock.ExpectExec("INSERT INTO structure_events").
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO workspace_auth_tokens").
WillReturnResult(sqlmock.NewResult(0, 1))
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
@@ -438,6 +440,9 @@ func TestWorkspaceCreate_SaaSHardForcesTier4(t *testing.T) {
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO structure_events").
WillReturnResult(sqlmock.NewResult(0, 1))
// External workspaces return early with connectionToken in the
// connection payload; they do NOT reach the inline auth_token mint
// at the bottom of Create (non-external path only).
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
@@ -478,6 +483,8 @@ func TestWorkspaceCreate_WithSecrets_Persists(t *testing.T) {
// canvas_layouts (non-fatal, outside tx)
mock.ExpectExec("INSERT INTO canvas_layouts").
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO workspace_auth_tokens").
WillReturnResult(sqlmock.NewResult(0, 1))
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
@@ -559,6 +566,8 @@ func TestWorkspaceCreate_EmptySecrets_OK(t *testing.T) {
mock.ExpectCommit()
mock.ExpectExec("INSERT INTO canvas_layouts").
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO workspace_auth_tokens").
WillReturnResult(sqlmock.NewResult(0, 1))
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
@@ -598,6 +607,7 @@ func TestWorkspaceCreate_ExternalURL_SSRFSafe(t *testing.T) {
mock.ExpectExec("UPDATE workspaces SET url").
WillReturnResult(sqlmock.NewResult(0, 1))
// CacheURL is non-fatal — uses Redis (db.RDB, set by setupTestRedis), not the DB.
// External workspaces return early before the inline auth_token mint.
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
@@ -1809,6 +1819,8 @@ runtime_config:
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO structure_events").
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO workspace_auth_tokens").
WillReturnResult(sqlmock.NewResult(0, 1))
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
@@ -1868,6 +1880,8 @@ model: anthropic:claude-sonnet-4-5
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO structure_events").
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO workspace_auth_tokens").
WillReturnResult(sqlmock.NewResult(0, 1))
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
@@ -1921,6 +1935,8 @@ runtime_config:
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO structure_events").
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO workspace_auth_tokens").
WillReturnResult(sqlmock.NewResult(0, 1))
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
@@ -2066,6 +2082,8 @@ func TestWorkspaceCreate_188_ExplicitRuntimeNoTemplate_OK(t *testing.T) {
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO structure_events").
WillReturnResult(sqlmock.NewResult(0, 1))
mock.ExpectExec("INSERT INTO workspace_auth_tokens").
WillReturnResult(sqlmock.NewResult(0, 1))
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)